DK Street Journal

Hertz is charging more for a day's rental than in any second quarter on record outside the 2022 travel boom, and its cars are still getting more expensive to own. Revenue per day reached $62.11 in the June quarter, up 9% year over year, on total utilisation of 79% from a fleet 1% smaller than a year earlier. Over the same three months, net depreciation per unit ran $302 a month against $256, an 18% increase that the company attributed in part to the non-recurrence of prior-year tariff-related benefits.

That $302 is not a price anyone paid. It is Hertz's own estimate of what a vehicle bought twelve to twenty-four months ago will be worth when it is sold, charged monthly and revised prospectively, which means one change in residual assumptions moves reported profit without a single additional rental. The June warning was widely read as a passing used-car blip. In September the Manheim Used Vehicle Value Index printed 205.9, 0.6% below a year earlier on an adjusted basis and the first year-on-year decline of 2026, and Cox Automotive cut its full-year forecast for the index from a 2% gain to roughly 0.2%.

The $4 that came from the auction lane, not the counter

Hertz's gross depreciation per unit was $298 for the quarter. The gap to the $302 net figure came, in the company's prepared remarks of 6 August, from "the loss on sale of a concentrated mix of older vehicles and pressure on the wholesale market dynamics that have since normalized." The same remarks said the fleet strategy "should continue to yield strong near-term results and solidify the ability to deliver net DPU below $300 per month," with about 94% of the US core fleet model year 2025 or 2026 — a younger, smaller book rebuilt after the electric-vehicle write-down cycle.

"We're staying focused on what we can control, our fleet, our cost, our revenue performance, and the customer experience we deliver," chief executive Gil West told investors on the August 6 call.

The cost of getting the estimate wrong was demonstrated in June, when Hertz guided net depreciation per unit to roughly $300 because used-car softness converted April gains on vehicle sales into May losses, and trimmed adjusted corporate EBITDA guidance to $50m–$80m. On 24 June it priced 37,037,037 shares at $2.70 to raise $100m alongside $300m of exchangeable notes; diluted shares went from 314m in March to 418m in June. A securities class action followed, alleging that recurring used-market weakness was materially depressing both metrics.

The shares have fallen 73.5% in six months, from $6.99 in mid-April to $1.85, and the downtrend began on 10 June, two weeks before the capital raise was announced.

A thin equity sliver is not a cheap enterprise

At $1.85 the equity is worth $584m. Trailing earnings are negative, consensus still has 2026 at -$0.77 a share, and book equity is negative, so the only workable anchor is enterprise value against EBITDA: 7.03x trailing, which is dearer than Avis Budget at 5.83x. The low share price reflects what is left over after the fleet debt, not a discount on the business.

Avis cut its cost per car and fell harder

Avis Budget, which rents under the Avis, Budget, Payless and Zipcar brands, ran the meter the other way. Per-unit fleet cost fell 4% to $290 a month excluding currency, revenue per day held flat at $67.84, and a 5% smaller average fleet lifted utilisation 1.9 points to a second-quarter record 72.6%. Its shares are down 74.5% in six months, but the starting point was a closing peak of $713.97 on 21 April, after which Fugazi Research alleged a manufactured squeeze built on holdings of more than 49% by SRS Investment Management and about 20% by Pentwater Capital; the stock lost 67.9% across two sessions. Measured from a year ago it is down 32.4%, with unit costs lower and utilisation at a record — the part of the decline the fundamentals do not account for.

Avis still carries the residual risk on about 84% of its fleet and $19,850m of vehicle-programme debt struck against those same values. A one-percent residual error across the book is roughly $200m by that arithmetic, about 5% of the $3.71bn equity.

Ryder's residuals have not turned

Ryder is barely a rental company in earnings terms. More than 90% of its revenue sits under long-term contracts, lease capital spending is guided to $1.9bn against $200m for rental, and used truck and tractor prices rose 6% and 3%, still above the residual estimates it books for depreciation. Its stock is up 30.5% over twelve months; the 11.7% three-month slide left the multiple almost untouched at 19.28x trailing against 20.37x in early May, and the 29 September downgrade from Baird's Dan Moore cited elevated expectations rather than residual values.

The lag is the thing to watch there. September's weakness was concentrated where Ryder's book lives: non-electric wholesale prices fell 1.7% while electric values rose, and with diesel at record prices the days' supply of diesel vehicles reached 39, up 27% year over year. A full-service lease signed three years ago was priced off an assumed Class 8 residual, and that assumption pays out only when the truck is sold.

Ryder reports on 22 October, the first of the three and the first look at whether used-truck proceeds held above booked residuals through the September decline. Hertz follows on 5 November, when the quarter that contains that decline finally shows up in the only number that matters at its counter: what a car cost to own while someone else was driving it.

nCino, which sells banks, credit unions and independent mortgage lenders the cloud software their loan officers and account-opening staff work in, expects to finish its fiscal year with more business under contract than it will record as revenue. Guidance issued on 25 August puts period-end annual contract value at $662.5m-$667.5m against full-year revenue of $644m-$647m. The gap is work that has been signed and not yet switched on.

That ordering is the whole difficulty in reading a vendor like this. The July quarter was taken as evidence that bank demand had re-accelerated, and in the recognised line it had: total revenue of $161.0m rose 8% and subscription revenue 10%, or 12% excluding US mortgage, where subscription revenue of $20.6m fell 1% as higher interest rates kept pressure on originations. But the revenue a bank-software company reports this year is what it sold in earlier years. What it is selling now shows up first in the contracted book.

The book is bigger than the guidance

That book is now $1.4bn. nCino's remaining performance obligations reached $1.4bn on 31 July, against $1.3bn at the end of January and $1.04bn two years earlier - roughly 35% growth over two years while reported revenue grew 8% to 10% a year. The company expects about 65% of it to convert to revenue within 24 months and 29% in the 25 to 48 months after that.

The gate between the two is implementation, and it is getting smaller. Professional services revenue was $17.5m in the July quarter, down 3% year on year, at a gross margin of 3% after a 600 basis-point improvement from negative 3%; services fell from 12.1% of revenue to 10.9%. Subscription net retention was 110%, or 106% organic. "We are seeing many of our largest customers consolidating more of their most critical operations on nCino and expanding their commitments to include our market leading AI capabilities," chief executive Sean Desmond said on 25 August.

Guidewire, which sells property-and-casualty insurers their core policy and claims systems, runs the same mechanic one cycle further along. "More than half of the net new ARR contemplated in this outlook is already under contract with ramp dates defined and signed customer agreements," chief financial officer Jeff Cooper told investors on 3 September; Guidewire closed its fiscal year with recurring revenue of $1.242bn, up 19% in constant currency. It is also the most expensive name of this group at 40.1x forward earnings.

Q2's meter is already running

Q2 Holdings supplies the digital-banking front end to regional and community institutions, and its forward book is converting rather than waiting. Subscription recurring revenue reached $825.5m, up 15%, with committed backlog up $404.3m year on year to about $2.8bn on revenue of $219.8m in the June quarter, up 13%. Gross margin went from 53.6% a year earlier to 59.2% and operating margin from 5.0% to 13.4%. The company raised full-year guidance to $881m-$886m and added a $350m repurchase authorization, with non-GAAP gross margin of 62.3% reflecting a completed cloud migration.

It also has a second source of growth that does not require winning a bank at all. Helix, its banking-as-a-service platform, serves about 18 million end users and moves roughly $40bn a year; transactional revenue was $17.9m against $16.7m a year earlier, even as discretionary services work declined.

BlackLine is cutting its own seat count on purpose

BlackLine sells accounting-close automation to corporate controllers, and its two meters point in opposite directions. Reported revenue growth has accelerated across four quarters to 9.2%, but recurring revenue of $719m grew only 6% and net revenue retention fell to 102.4% from roughly 105%. The reason is a decision rather than lost demand: platform pricing, which replaces per-seat licenses with unlimited users and gates access to its Verity AI agents, reached 17% of eligible recurring revenue from 13%, took nearly 90% of net new business, and is expected to shrink paid user counts.

Chief executive Owen Ryan put the bet to investors on 4 August: "Gartner expects the average Fortune 500 company to be running more than 150,000 AI agents by 2028, up from fewer than 15 last year, and fewer than one in five companies believe they have the governance to manage that scale."

The prices moved together; the meters did not

All three sit inside a year-long hole that began as a verdict on the software model, not on these businesses: roughly $2trn of software market value went after a 6 February selloff triggered by fears that AI agents would bypass per-seat subscriptions. Over ninety days to 12 October nCino is up 12.6% and Q2 up 13.9%; over twelve months nCino is down 21.1%, Q2 down 5.3%, and BlackLine down 39.6%. BlackLine has risen 4.7% over thirty days off a 29 September low.

On price to gross profit, nCino sits at 5.65x trailing against 5.52x in May - essentially unchanged while its GAAP operating margin swung from minus 6.2% to plus 8.5% year on year. Q2 carries the premium at 7.64x, about 35% above nCino, having risen from 7.28x in May. BlackLine has compressed to 3.15x from 3.58x, with a 9.8% free-cash-flow yield.

So the three-month advance at nCino and Q2 is supported by what the accounts show, and the twelve-month decline at nCino is not: 9% growth with a contracted book growing faster, forward earnings at 15.1x against consensus for 45% EPS growth, and a 6.1% cash yield do not describe a business that lost a fifth of its value. The split has a cause in each case. Q2 is rewarded because its growth is already in revenue and its migration spending is behind it. nCino is not, because its evidence of winning is a disclosure line rather than an income-statement line. BlackLine's de-rating tracks a real slowdown in the forward meter that its accelerating revenue hides - and at this cash yield, it looks closer to finished than to beginning.

The risk in all three books is the customer disappearing rather than refusing to renew. More than 130 bank deals and roughly 100 credit-union mergers were under way in 2026, after 180-plus in 2025, with technology and compliance cost behaving like a fixed charge that lands hardest on the smallest balance sheet. Loans at commercial banks grew 5.1% year on year with the community-bank net interest margin at 3.81% - enough to fund a software renewal, not enough to make the buyer indifferent to its price.

BlackLine reports after the close on 5 November, and it is the one of the three whose next print can settle its own question: whether taking seats off the bill was the price of selling agents, or just the beginning of a smaller bill.

The rating arrived, and it went against the company

Alignment Healthcare, a Medicare Advantage insurer that runs its own health plans in California, North Carolina and Nevada, lost the fourth star on its largest contract. The Centers for Medicare and Medicaid Services published the 2027 star ratings on 8 October, and the company disclosed the same day that California HMO contract H3815, which serves approximately 75% of its health-plan membership, expects 3.5 stars for 2027 against 4.0 for 2026. Six other eligible contracts held four stars or better, three of them at 4.5. The share of Alignment's members sitting in four-star-or-better plans next year falls to roughly 25% from about 98%.

Through the summer Alignment was read as evidence that a small capitated insurer can out-execute the majors while they retreat: it raised full-year guidance on 30 July with the best cost ratio in its public life. The file contradicts none of that. It removes money from a line none of those meters touch, and removes it in 2028.

What a star is worth, and when it pays

The rating is a price lever. In most counties a plan at four stars or better has its county benchmark raised by five percentage points, ten in designated double-bonus counties, and the rebate percentage, the share of the gap between bid and benchmark a plan keeps to fund supplemental benefits, steps from 50% below 3.5 stars to 65% and then 70%. H3815's half-step forfeits the benchmark bonus while staying inside the 65% rebate band. Industry-wide, quality bonuses run about $13bn a year, and the 2027 ratings govern what is paid in 2028, which is why Alignment can say it anticipates no impact on 2026 or 2027 revenue and be right. TD Cowen put the annual cost of the lost star at roughly $175m, against the $96.7m of EBITDA consensus expects for all of 2026. That single comparison is the case for the de-rating.

The accounts were going the other way

June-quarter revenue was $1,335.6m, up 31.6%, with gross profit up 35.5% and operating income up 85.1%; the adjusted medical benefit ratio came in at 86.3%, forty basis points better than a year earlier. Derived from reported revenue and the disclosed 294,100 members, revenue per member per month was about $1,514 against roughly $1,507 a year earlier. The growth line is enrolment, almost none of it price, which is precisely why a badge matters: the contract carrying three-quarters of the members will be shopped at 3.5 stars when open enrollment starts on 15 October.

The next print is guided lower by design, to $1.30–1.32bn of revenue and $20–30m of adjusted EBITDA on a seasonally higher cost ratio. At Baird's healthcare conference on 15 September management named more aggressive hospital billing, heavier disputes and appeals and longer skilled-nursing stays, and disclosed $10–11m of unplanned second-half spending; the shares fell 19.9% that session. "Whenever you go through a claims application implementation, those are messy and noisy," chief executive John Kao said that day. A whistleblower suit disclosed in early July, alleging about $8–10m of operating expenses were booked as capital expenditure in 2024, remains unresolved; the company calls the claims meritless.

Clover's fifth star is sitting in an appeals court

Clover Health, which sells PPO and HMO Medicare Advantage plans built around physician software, was awarded 5 stars on its PPO plans and 4.5 on the HMO, with the PPO carrying about 97% of its members and a HEDIS score of 4.82 out of 5. "Success in the Star ratings system means an enhanced ability to support our members and to continue to provide them market leading plans," chief executive Andrew Toy said on 8 October. The top rating is conditional: CMS calculated it excluding the twenty measures covered by a 27 May 2026 district-court decision in Clover's favour, which is on appeal, and with those measures included the PPO would have been 4.5.

The underlying book is one business, not three. Insurance revenue of $737.8m was 99.3% of the $743.2m June-quarter total, the insurance benefit expense ratio improved eighty basis points to 87.6%, and gross profit per member works out near $324 a month on 157,309 members, against roughly $205 at Alignment on a book nearly twice the size. Clover trades at 4.47x forward price-to-gross-profit; Alignment is at 2.44x, having started September at 4.13x.

What the file settles and what it leaves open

None of Alignment's decline is explained by its reported results, which improved on every meter it discloses. The lost bonus dollar is nonetheless larger than this year's expected EBITDA and will not appear in a financial statement for eight quarters, so both readings of the price are defensible, and the sell side has written both at once: JPMorgan cut its target from $22 to $10 while keeping its Overweight rating, KeyCorp from $28 to $12. The real question is sequencing: whether the appeals and the measures Alignment says it intends to litigate restore the star before the 2028 payment year, and whether a selling season spent at 3.5 stars slows the enrolment engine that is the whole revenue story. The company says it remains confident of returning the California contract to at least four stars.

Seniors start choosing on 15 October from a page that shows one number per plan. Whatever the lawyers eventually produce, it will not change the number on that page this season.

Navitas Semiconductor closed its purchase of Claros on 6 October, paying up to $232.8m — about $216.0m at closing, roughly $125m of it cash — for the ability to put a voltage regulator directly beneath a high-current processor. That capability, integrated voltage regulation and vertical power delivery, is the last stage in an AI rack's power chain, and it is the stage Monolithic Power Systems already supplies.

The market spent September deciding Monolithic was losing it. A research note questioning the company's allocation on Nvidia's next platform cost the shares 7.39% in the 14 September session, from $1,234.46 to $1,143.24 on 978,292 shares against the 68,339 traded six sessions earlier. The filings published six weeks before that say nothing of the sort about the business as it stands. The reason both can be true is structural: a power-semiconductor dollar is earned in a way the revenue line cannot read back.

The dollar arrives late, and through someone else

Monolithic, a Kirkland, Washington designer of DC-to-DC converters that step voltage down for servers, cars and industrial gear, is paid for content on a board that was designed years before it shipped. It is also paid at one remove: 88% of sales in both the three and six months to 30 June 2026 ran through distribution arrangements whose enforceable rights and obligations are with distributors and resellers, not end customers, and two of those distributors were 28% and 15% of revenue. Reported revenue is, to that extent, a distributor ordering pattern.

What it reported was acceleration. June-quarter revenue of $980.6m grew 47.6% year on year, the fourth straight quarter of faster growth, and Enterprise Data revenue rose 164.3% to $380.6m, lifting that end market to 38.8% of the company from 32.7% a quarter earlier. Operating margin went to 31.0% from 24.8%. GAAP gross margin has sat within 23 basis points of 55.2% across four quarters. Days of inventory fell 17, to 140.

On the 30 July call, management said channel inventory "remains very low" and book-to-bill was "well above 1.0x." Founder-chief executive Michael Hsing raised the year's floor for the segment: "I think right now, based on what we can see, we're willing to raise the floor for that particular end market from 85% for the year to 130% for the year," he told investors.

The number nobody discloses

The quarterly release breaks out revenue by end market against a single consolidated gross margin. There is no margin by segment, and there is no disclosure anywhere of how many of the power phases on a given accelerator board are Monolithic's rather than a second source's. That is the quantity in dispute. Rosenblatt, cutting the stock to Neutral, allowed that Monolithic is "capturing the pole position in power delivery to accelerator sockets" while arguing Nvidia's roadmap looks to have Texas Instruments and perhaps Analog Devices in the Rubin cycle. Its $880 target, set on an assumed $22 of fiscal 2026 earnings per share, now sits far below both the $1,386.29 price and the $27.45 consensus — a stale price view carrying a live competitive observation.

By the time a lost phase reaches the income statement, the platform carrying it has already shipped.

The other side of the socket

Navitas is what the same end market looks like before the revenue arrives. June-quarter revenue of $10.5m fell 27.3% year on year; GAAP gross margin was 0.4% including about $4m of intangible amortisation, 39.5% on a non-GAAP basis, and the non-GAAP operating loss was $11.4m. Diluted shares rose 18.6% in twelve months, and the company ended June with $557.4m of cash after raising roughly $373m in the quarter — the cash half of the Claros price consumes about a fifth of that. Chief executive Chris Allexandre has said AI infrastructure will exceed a third of revenue by year-end, which is a real share of a $36.5m trailing base priced at 77 times sales. The 800-volt in-rack date moved to early 2028 at Citi's conference on 8 September, from 2027 on the July call, and the joint 800-volt reference design with Microchip unveiled 5 October is a demonstration rather than a volume agreement.

Analog Devices shows the third route: buy in and carry it on a diversified book. It paid $1.5bn for Empower Semiconductor, grew August-quarter revenue 39.6% to $4.022bn at a 67.3% gross margin, and trades at 31.6 times forward earnings against Monolithic's 50.5 times.

What September was arguing about

Monolithic's shares are 18% below their 3 June peak of $1,689.89, flat over three months, and up 14.7% in thirty days inside a sector-wide advance — the iShares Semiconductor ETF gained 11% in September. The results account for essentially none of the drawdown and the valuation accounts for most of it: 84.5 times trailing earnings and 37.7 times trailing gross profit, against the roughly 39 times gross profit that already sat at the top of the range in August. The dispute is about the price, not the reported business.

What the dispute cannot be settled by is another quarter like the last one. On 29 October the company reports a period already guided to $1,140m–$1,160m of revenue, roughly a sixth above June, at a gross margin of 55.2%–55.8%. The figure that decides the next three years is not in that release, and will not be in any release until the boards it describes are already in racks.

EMCOR Group, the US and UK specialty contractor that designs, installs and commissions the electrical and mechanical systems inside buildings, finished June with more work under contract than at any point in its history and raised its profit forecast for the year. Its shares have gone nowhere since the summer and sit 16% below their 2026 high.

The layer of the artificial-intelligence buildout that actually pulls the conduit has been sold off since industry executives publicly called for the industry to slow its own spending, which Reuters described as investors repricing AI capital-expenditure risk across the supply chain. A contractor's revenue line cannot settle whether that worry is right. Under percentage-of-completion accounting, revenue is cost incurred to date measured against management's estimate of a job's total cost, so reported revenue is a progress estimate and reported margin is a forecast, revised every quarter, with jobs signed two years ago landing in the current period.

The order book grew faster than the work burned off

Remaining performance obligations, meaning contracted work not yet performed, reached $17.14bn at 30 June, up 43.9% from a year earlier, with the largest gains in network and communications, water and wastewater, and institutional and healthcare work. That balance was $13.25bn entering the year and $15.62bn at the end of March. Set the $1.52bn quarterly increase against the $5.155bn of revenue recognized, and the implied book-to-bill, derived from those two balances, is roughly 1.3x. Operating margin widened by about one percentage point to 10.6% on 19.8% revenue growth, and guidance went to $32.00–33.25 of diluted earnings a share on revenue of $20.00–20.50bn.

The buildout's effect on the work itself is scope, not square footage. An AI data center carries a content "multiplier of 1.5 to 2" for mechanical systems and "probably 1.5" for electrical work against a conventional facility, chief executive Anthony Guzzi told investors on 30 July. Electrical construction revenue in the network and communications sector rose 45% in the quarter.

Where the customer's cash actually goes

EMCOR holds $2.609bn of contract liabilities, amounts invoiced and advanced ahead of revenue recognized, against $377.3m of contract assets at the end of March, per its quarterly filing. That float is not an earnings engine: net interest income was $4.322m in the June quarter, under 1% of the $547.3m of operating income, and cash fell from $1.112bn to $924.4m over the half as the customer money financed unbilled work and growth.

What would settle the margin question is the split between fixed-price jobs, where labor inflation and an unpriced change order are the contractor's loss, and cost-reimbursable work, where they are the customer's. EMCOR's annual report describes that risk allocation without quantifying the mix.

The one meter that would settle it is not in the filings.

The exposure is not small in principle: electrical work runs 45% to 70% of a data center's construction cost, and construction labor costs in primary North American data-center markets rose 8% to 12% a year in 2024-25. Microsoft president Brad Smith has called the electrician shortage the "#1 problem" for US data-center expansion. EMCOR's answer is union labor mobility and prefabrication, and the trace shows in output per head: FY2025 revenue across 44,000 employees works out near $386,000 each, and the 2026 guidance midpoint near $460,000, a derived 19% gain without more bodies.

Argan is the undiversified version of the same accounting

Argan, a lump-sum engineering and construction firm that builds whole gas-fired power plants, is where estimate risk has already bitten. Consolidated gross margin stepped down from 25.0% to 21.0% to 19.3% over three quarters, and the industrial segment fell to a 7.3% margin on two projects with higher estimates-to-complete than initially estimated. "As projects ramp and get into their second and third year of construction, we expect to see higher revenues and with successful execution, we have the opportunity to enhance margin," chief financial officer Joshua Baugher said on 2 September. Backlog fell from $2.9bn to $2.5bn, 80% of it four US gas plants. Here the float does flatter earnings: FY2026 net income of $137.8m exceeded operating income of $134.7m despite tax, against cash and investments of $895.0m.

Dycom, the telecom-infrastructure contractor, is barely on the same meter. It is paid per unit of work under master service agreements, which were 76.9% of contract revenues in fiscal 2022, and AT&T, Verizon with Frontier, and Lumen together supplied about half of fiscal 2026 revenue. Its backlog rose 53.2% to $12.242bn, of which more than $1bn is fiber tied to data centers, about 8% of the book.

Three names, one shelf, three different reasons for the price

Over the three months to 9 October, EMCOR was up 0.8% while Argan fell 42.9% and Dycom 37.0%. Only Argan's decline has a reported business change behind it: a margin trajectory down three quarters, a backlog that shrank, and a year carried by a handful of jobs against a company capacity management puts at 10 to 12 at once. At 29.7x forward earnings and 23.6x trailing enterprise value to EBITDA after a 51% fall from its high, that reads as a justified re-rating. EMCOR's own multiple has compressed from 31.4x trailing in early May to 24.6x, against Comfort Systems at 42.2x trailing and 35.0x forward, a gap part-explained by Comfort's 50.3% June-quarter revenue growth. Dycom trades at 15.8x forward against 24.3x trailing. For EMCOR and Dycom the order books and guidance moved the other way from the shares, and the likelier reading is the group being repriced as one block; what cannot be checked from outside is whether EMCOR's growing book is taking on fixed-price risk as it goes.

That is the uncomfortable part of a business whose revenue is an estimate. EMCOR reports third-quarter results on 29 October, and the order book will be printed to the decimal. The figure that decides what it is worth will not be.

Newmont told investors in July that producing an ounce of gold had cost it $1,621 in the June quarter, against $1,029 three months before. Nothing in the mine plan of a company that digs gold in ten countries changed by 58% in ninety days. The step came from the timing of sustaining capital, from by-product credits that had flattered the earlier figure, and from the incremental costs of running Cadia in New South Wales through the downtime that followed a magnitude 4.5 earthquake there on 14 April.

All-in sustaining cost is the figure the gold industry publishes as its margin meter, and revenue is where investors look for the metal price passing through. For a producer this size, neither line does the job it is asked to do. Newmont's realised price was $4,900 an ounce in the first quarter and $4,414 in the second; set against the two cost figures, the spread per ounce narrowed 27.9% while reported revenue fell 16.3%. The income statement understated the squeeze by roughly twelve percentage points, and the cost line did marginally more of the damage than the weaker gold price.

What the cost per ounce leaves on the floor

The measure is defined to exclude the next mine. Agnico Eagle, which mines in Canada, Mexico and Finland, reported total cash costs of $1,054 an ounce and all-in sustaining cost of $1,459 in the June quarter: the $405 difference is sustaining capital and related charges. Growth capital sits outside both. Newmont plans $1.4bn of development capital this year and spent $524m of it in the first half, which against guidance of 5.26m to 5.3m attributable ounces works out at roughly $265 an ounce of spending the cost line never shows.

Which is why the cost step and the cash machine can run in opposite directions. Newmont still generated record free cash flow of $2.2bn in the June quarter, after $3.1bn and a raised $6bn repurchase authorisation in the March quarter. The shares trade at 12.6x forward earnings against 14.8x trailing, with a 7.8% trailing free-cash-flow yield.

The $580 an ounce that geography costs

AngloGold Ashanti, which mines in Africa, Australia and the Americas and reports in dollars, spent $2,039 an ounce in the same quarter, up 22.4% year on year, with sustaining capital of $332m against $273m. Of the increase in its cash costs, the company attributed $216 an ounce to macroeconomic factors and $29 to operational ones. "Royalties, fuel, broad inflation, FX basically accounted for all of the increase," chief executive Alberto Calderon said on the results call; chief financial officer Gillian Doran told the same 31 July call that "the upward pressure on costs for the industry were particularly acute this quarter." A fatality at Obuasi on 24 April cost two weeks of operation, and the mine has been running without its ore-pass system after a catastrophic failure.

Agnico's ounce came out of record throughput: the Detour Lake mill ran at 80,275 tonnes a day and produced 207,279 ounces, up 23.2% year on year. The company has hedged about 60% of its remaining 2026 Canadian-dollar exposure in a 1.38 to 1.42 collar. That cost base is what the market is paying for, at 16.1x forward earnings against 16.2x trailing and the group's thinnest free-cash-flow yield at 4.7%, while the soft-currency structure in Ghana and Tanzania is priced at 10.6x forward against 12.5x trailing and an 8.9% yield.

Why last month's slide was mostly the metal

The majors fell together over the thirty days to 9 October, Agnico 7.3%, Barrick 7.9%, Newmont 8.7%, after the Federal Reserve raised rates on 16 September, the ten-year Treasury yield reached 5.3% and gold dropped more than 6% in the month. That is close to one-for-one with the metal, less than the leverage gold equities usually deliver, and it is not a cost story. Barrick's jurisdiction overhang is behind it: Mali was settled for $430m in November 2025, and its shares trade at 11.8x forward against 10.6x trailing, the only one of the four whose forward figure sits above trailing.

What moves a gold equity hard is an ounce that does not arrive. Kinross fell 21.8% over the month after cutting production guidance on 24 September to two to three per cent below its earlier low end, citing storms at La Coipa and lower grades and recoveries at Round Mountain.

So the drawdown belongs to the rate shock, and the three-month gains of 24.3% at Newmont and 26.8% at Agnico belong to a metal that, industry-wide, outran its cost base: average producer sustaining costs hit a record $1,785 an ounce in the March quarter, the 28th consecutive annual increase, and the margin over them hit a record anyway. The business underneath is slowing. Revenue growth halved sequentially at all three, gross margins compressed between four and nine percentage points, and sustaining cost per ounce rose 13.9% year on year at Agnico. Newmont's management said on 23 July that sustaining capital would be approximately 58% weighted to the second half, citing programmes at Boddington and Cadia, ventilation at Tanami and summer construction in British Columbia. The cost line for the September quarter is already largely written.

Spot gold was about $4,184 an ounce on 12 October, a quarter below January's record. The price resets every morning; the cost of the next ounce was fixed by a capital schedule set in the spring.

In the June quarter Rambus sold more memory-interface silicon than it collected in patent royalties for the first time, and then pointed the two lines in opposite directions. The San Jose company, which sells register clock drivers and power-management chips designed into DDR5 server memory modules and licenses memory and security cores to chip designers, guided September-quarter product revenue to $110-116m and royalties to $69-75m, up and down roughly 14% sequentially.

A record quarter was read as the AI memory build arriving at the company paid whenever somebody else's module ships. The four quarters to June tell a narrower story: revenue of $756.3m, up 17.2% on the prior four, carried summed net income up only 4.6%. Gross margin widened to 79.8%, and operating income still grew slower than revenue in each of the last two quarters. The mix is doing that, because the dollar supplying the growth is per-module silicon and the dollar it is displacing is a near-fully-margined annuity.

The annuity is now billed as fast as it is booked

Fixed-fee patent licensing is the line where reported revenue can mislead, since recognition is front-loaded at signing. That defense has stopped applying. June billings of $84.1m sat against $84.2m of recognized royalties; March was $70.8m billed against $69.6m recognized. Unbilled receivables of $27.8m and deferred revenue of $25.7m are small beside a royalty run-rate near $290m a year. The royalty line is reporting demand, and the guide takes it down.

The content step belongs to Intel's and AMD's calendars

Chief executive Luc Seraphin credited "the migration from 12 to 16 memory channels per processor" for the June outperformance on the 27 July call. That is the right meter: module units times dollar content, not bit prices. Micron's August-quarter margins and rising contract DRAM prices pass Rambus by entirely. Registered-module bit supply is growing 15-20% while high-bandwidth memory absorbs a rising share of the top three suppliers' wafers, about 22% this year and 30% next; Baird cut its 2027 module unit growth estimate to 12-15%. Management said it saw no pricing opportunity in the shortage.

The upgrade cycle is real and late. Rambus puts the multiplexed-rank module market at roughly $600m, with initial volumes late in 2026 and material revenue in 2027; Intel adds second-generation support to its Xeon 6900P platform in the first quarter of 2027. The socket is contested by Renesas, at roughly 35% of the register-clock-driver market, and by Montage Technology. And the memory-pooling promise the category is named for stays in the IP business: management called merchant Compute Express Link silicon a fragmented, ASIC-like market it addresses by licensing, with silicon IP guided to 10-15% annual growth.

A re-rating with no report in between

The shares closed at $107.47 on 9 October, up 22.6% in thirty days, which lifted price-to-trailing-gross-profit to 19.69x from 16.06x a month earlier and back within 6% of the 20.92x struck in early May. Gross profit did grow 28% over that stretch, so part of the move is earned. The rest is sector re-risking after a selloff that cost a semiconductor basket roughly $2.1 trillion from the late-June peak, with Argus raising its target to $123 on 23 September the only company-specific news in the window. Rambus remains 37% below its 3 June close of $170.66.

So the results account for the direction of the advance and not its level. A multiple from May has been restored to a business whose highest-margin line is guided lower, whose unit meter is being marked down by the people who forecast module shipments, and whose content step arrives with 2027 server platforms. Forward estimates still model revenue growth near 18% this year and 20% next, with earnings per share growing faster than sales, which requires the operating leverage that the last two quarters did not deliver.

Rambus reports third-quarter results on 26 October at 2:00 p.m. Pacific. It is the first look at whether a flat annuity and a growing chip business still add up to a company earning more than it did a year ago.

Dollar Tree's summer gross margin was the widest the chain has printed in years, and four-fifths of the improvement was a refund of duties the Supreme Court decided the government never had the authority to collect. After February's ruling that the International Emergency Economic Powers Act did not support the tariffs levied under it, roughly $166bn of duties began returning to importers of record, with interest, through a Customs and Border Protection refund portal. Dollar General and Ollie's Bargain Outlet booked their shares inside cost of goods sold in the same quarter.

The bridges were read as proof that the trade-down shopper is finally paying off at the margin line. All three companies are instead handing the refund to that shopper as lower shelf prices, which makes the expansion a one-year event; and underneath it, only one of the three businesses got better.

The refund, by how much of the margin it bought

At Dollar Tree, which sells a fixed-price assortment of food, household basics and seasonal general merchandise, 680 of 850 basis points of gross-margin expansion in the quarter ended 1 August was net tariff impact, taking gross margin to 42.9%. At Ollie's, the closeout chain that buys brand-name overstock and liquidation lots, the refund was worth 380 basis points of a 360 basis-point expansion: all of it, and more. At Dollar General, the rural small-box chain, 81 of 127 basis points.

None of the three banked it. "We will not hesitate to invest beyond this level to strengthen our price leadership position," Eric van der Valk, Ollie's president and chief executive, told investors on 2 September. Dollar Tree is putting about $210m of its $383m back into cost of goods, marketing and store operations.

Dollar Tree's price is printed before the cost is known

Dollar Tree publishes its price in advance, so landed cost, freight and duty land in gross margin rather than in revenue. The levers are the item behind the price and the share of the assortment lifted above the $1.25 anchor: multi-price merchandise reached 17% of total sales, up about 400 basis points year over year, across roughly 6,600 stores. "Rather than passing those costs on to the customer, we've taken advantage of the fact that we're getting that lower tariff rate in and that tariff rate is absorbing inflation," chief financial officer Stewart Glendinning said on 27 August.

That absorption has a term. Management guided third-quarter gross margin flat and fourth-quarter margin down year over year on fuel, freight and merchandise inflation, with third-quarter adjusted earnings of $0.80 to $0.95 a share including a $0.50 reinvestment headwind. Investors priced the guide: the shares fell through late September and are 16.3% below the 25 August high of $133.74. At 15.0x forward earnings against 14.5x trailing, the forward figure sits higher because consensus of $7.90 is below the $8.19 of trailing earnings that contains the refund quarter's $1.31.

Ollie's buying was never the problem

Ollie's margin is a buying outcome, and the buying was fine: TJX, Ross and Burlington all describe closeout supply as highly favorable. Demand was the problem. Comparable sales fell 1.8% on flat traffic and a smaller basket, with unseasonable weather in lawn and garden and air conditioners costing more than 100 basis points of comp; acquired Big Lots boxes are comping low-to-mid single digits negative in their second year. Full-year sales guidance was cut to $2.928–2.941bn from $2.98–3.00bn, comps to flat-to-+0.5%.

The stock then rose 17.6% in thirty days, alongside TJX at 9.9% and Burlington at 13.7%, while Five Below fell 13.9%. The likelier reading is a bid for closeout and off-price retailing generally; consensus earnings of $4.61 sit at the midpoint of the cut guide, with no upward revision behind the move. At 18.5x forward Ollie's is the dearest of the three, and 19.0x trailing against 36.7x a year ago is a de-rating interrupted.

Dollar General earns it on shrink and trucks

With consumables at 82% of net sales, Dollar General has no price point to defend and makes its money on shrink, markdowns and distribution cost. "This marks the fifth consecutive quarter of growth in customer traffic," chief executive Todd Vasos said on 27 August, on comps of 3.5% with traffic up 2.0%. Operating income rose 29.2% on sales up 5.2%, full-year earnings guidance went to $7.80–8.00 from $7.20–7.45, and up to $700m of buybacks was pulled into the second half. The shares are up 28.9% over twelve months while the trailing multiple fell to 16.5x from roughly 18.3x.

What the pump price decides

The refund accounts for four-fifths of Dollar Tree's margin gain, roughly two-thirds of Dollar General's and all of Ollie's. What it does not explain is why the traffic went one way. Gasoline at $4.359 a gallon, up 39.4% in a year, taxes a destination trip and spares a fill-in one: Ollie's blames longer drive times in Midwest, Texas and Western markets, while Dollar General's shopper is a few miles away and its remodel program adds comp mechanically. There is also a claim on the windfall itself, since retailers that passed the duties to shoppers now face consumer class actions arguing the refunds belong to those shoppers.

All three report again in December. By then the refund is spent money, and what is left on the shelf is a price each company has promised to hold without it.

Meta launched a consumer artificial-intelligence assistant called Muse in late September, and among the shares that fell hardest were those of a company whose largest earnings engine is holding other people's idle cash. Charles Schwab, the retail brokerage and custodian for independent advisers that also owns banks, dropped 6.1% on 22 September to close at $100.35, on 16.9m shares against a typical 5m to 9m, its heaviest-volume down day of the quarter. LPL Financial fell 7% and Raymond James 3.5% on the same reasoning; the Nasdaq Composite rose 0.45% that day.

The trade priced a belief — that an assistant managing a household's money erodes demand for paid advice. Schwab's own revenue decomposition says that is the thinnest part of the franchise. Managed investing solutions produced $707m of the record $7.07bn of net revenues reported for the June quarter, about a tenth. Net interest revenue was $3.4bn, 47% of the total and up 19% on the year.

The line that got sold is not the line that pays

That $3.4bn is a banking spread wearing a brokerage's clothes. Uninvested client balances are swept off the brokerage statement into Schwab's own banks, reinvested, and paid back to the client at a rate Schwab sets. Transactional sweep cash ended June at $485.7bn, roughly 3.7% of $13.08trn in client assets, with on-balance-sheet bank deposits of $249.7bn and a further $70.6bn parked at third-party banks, where fees rose 35% on improved net yield. Margin loans add a second spread, at $165.1bn of client balances.

Six days before Muse, the Federal Reserve raised its target range 25bp to 3.75-4.00% in a unanimous vote, its first increase since 2023. A higher funds rate lifts the reinvestment yield on swept cash while the client rate stays administered. Net interest margin had already widened 12 basis points in the quarter to 3.00%, with management guiding fourth-quarter margin to 3.25-3.30%.

One caution the pitch usually omits: the cheap part of that repair is done. Higher-cost bank supplemental funding fell $44.8bn during 2025 to $5.1bn, so further expansion has to come from securities repricing and the policy rate. Asset gathering, meanwhile, accelerated: $120bn of core net new assets in the quarter, then a record August gather of $64.8bn, up 46% year on year.

Schwab lost 9.9% over the thirty sessions to 9 October, to $96.70. Interactive Brokers lost 3.8% in the same stretch and an S&P 500 index fund rose 2.2%, so the de-rating was Schwab's own.

Robinhood is paid rebates — and so is Schwab

Robinhood Markets, the 2,900-employee US platform for stocks, options, gold and crypto, is the supposed mirror: order-flow rebates instead of commissions. The mirror is cloudy. Schwab booked $528m of commissions and $624m of order-flow revenue in the June quarter, and that order-flow line alone exceeded Robinhood's entire $342m options business.

Robinhood's net revenues rose 32% to $1.31bn with net deposits of a record $21.7bn, while crypto revenue fell 38% to $100m and event contracts — regulated betting on outcomes — rose more than tenfold to $156m. Its cash spread is the laggard: net interest revenues grew only 9% to $389m, held back by lower short-term rates and weaker securities lending. Where the two firms differ is deliberate. Schwab chief executive Rick Wurster told the 21 July call the firm would support financial-event and economic-data products but has "no interest" in sports gambling or non-financial betting.

Robinhood fell 6.7% over the same thirty sessions, as bitcoin broke below $81,000 on 8 October and crypto liquidations approached $1bn. At 49.6x forward earnings against 47.2x trailing, consensus has FY26 earnings up 7.7% after 2025's 114%.

Nu's benchmark sits in Brasília

Nu Holdings, the São Paulo digital bank with 139m customers, shares the sector label and none of the plumbing. Its cost of deposits was 88% of the Brazilian interbank CDI rate, three percentage points better than a year earlier, funding an unsecured card and lending book: net interest income of $3.7bn and margin up 180 basis points to 22.9%. Copom cut the Selic to 13.75% on 16 September, the same week the Fed hiked. The blemish is credit — 90-day-plus non-performing loans rose 35 basis points to 6.9%. Nu rose 6.8% over the thirty sessions, trades at 18.5x forward against 21.8x trailing on a 33% return on equity, and reported-dollar revenue growth of 49.9% against 39% excluding currency, so roughly ten points of the headline was the real.

Which anchor you pick decides the answer

The business results account for none of Schwab's fall and arguably argue the other way. What they cannot settle is value, because Schwab's two anchors disagree. On earnings it is 14.9x forward against 17.5x trailing, the gap implying consensus FY26 earnings per share of $6.49, a third above FY25. On bank capital it is near 6.9x tangible book value of $14.10 a share, against a ten-year median of 5.29 — a premium, not a discount. Sell-side opinion split accordingly: Piper Sandler cut its target to $100 on 2 October while Bank of America raised its to $95 and stayed at Underperform.

The three names did not move together because they are not paid alike: Schwab on an administered deposit rate the Fed just made more valuable, Robinhood on volatility and product launches, Nu on a discount to a falling Brazilian benchmark.

Schwab reports third-quarter results before the open on 15 October, with consensus looking for $1.67 a share. The number that matters is the margin on cash, which is the one line Meta's assistant cannot touch.

China's tutoring companies are paid before they teach. Parents settle the term at enrollment, the money sits as a liability, and it turns into revenue class hour by class hour as the course is delivered. That makes the prepaid balance the only figure in these accounts that moves ahead of the income statement, and at TAL Education, the Beijing provider of after-school small-group classes under the Xueersi brands, it grew 25.9% in the year to 31 May, to $1.22bn, while the revenue it funds grew 31.9%.

The past quarter has been read as a recovery still gathering force: the three listed operators of this business left July downtrends behind and rose about 31% equal-weighted over three months. Their collections say the fastest point is behind them. New Oriental's deferred revenue rose 14.8% against 23.2% revenue growth; Gaotu's 18.9% against 20.2%. At each one the reported line is draining a book that was filled at a quicker rate, and both companies that gave guidance pointed at the high teens.

The regulator decides how far the money can run ahead

A framework issued by 13 departments including the Ministry of Education bars non-academic tutoring institutions from collecting a lump sum covering more than three months, 60 class hours or RMB5,000, about $716, and requires offline classes to finish by 20:30. So the forward book cannot be a year deep. Derived from the filings, TAL's balance equals 1.61 times the $758.4m it recognized in the May quarter, New Oriental's 1.46 times and Gaotu's 1.56 times: the same cushion of roughly a quarter and a half at three companies of very different size. The cash timing inside the year is violent. TAL took in $478.2m of operating cash in the May quarter, 63% of that quarter's revenue, after using $215.0m in the February quarter.

TAL filled capacity it had already leased

TAL's operating margin reached 18.1% in the May quarter against 2.5% a year earlier, on a gross margin of 57.8%, the best in five quarters. Centers are leased and staffed before seats are sold — more than 600 across 44 cities — so that margin is a utilization reading. "Our first quarter performance in fiscal year 2027 reflects solid execution across our businesses," president and chief financial officer Alex Peng said on 30 July. Two qualifications: revenue rose 32% in dollars but only 25% in renminbi, and the learning-content slab of books, apps and devices reached 35.3% of fiscal 2026 revenue from 31.8%. Yet group gross margin widened to 55.4% from 53.3%, so the hardware mix is not visibly diluting the service margin. The shares trade at 9.1x forward earnings, with price-to-gross-profit compressing to 3.73x forward from 4.37x trailing.

New Oriental's grocery arm is taking the margin

New Oriental's revenue growth accelerated through fiscal 2026 — 6.1%, 14.9%, 19.8%, then 23.2% — and non-GAAP operating margin reached 7.5%, up 470 basis points. Here the mix dilution is real: East Buy, the consolidated livestreaming and private-label goods business, grew 34.4% to $806.9m, 14.3% of sales, while group gross margin slipped to 54.6% from 55.4%. Overseas test preparation, the product the 2021 academic ban left untouched, grew only 6%; adults and university students grew 30%; non-academic tutoring booked 1,072,000 enrollments. "In this fiscal quarter, we executed cautious capacity expansion while carefully balancing revenue growth and operational efficiency," chief executive Chenggang Zhou said on 29 July. At 19.6x trailing earnings against 13.3x forward, the spread is what is being paid for that 14–18% guide, alongside a commitment to return no less than half of prior-year net income.

Gaotu buys its students at a price its classroom cannot cover

Gaotu's selling expenses were CNY913.2m in the June quarter, 54.7% of revenue, against a 66.5% gross margin: acquisition consumes about four-fifths of each gross-profit yuan, which is how that gross margin still ends at minus 9.0% at the operating line. Revenue growth has run 30.7%, 21.4%, 12.5% and 19.9% over four quarters, with the September quarter guided to 16.4–17.7%. The first half of 2026 carries a CNY101.2m net loss against a consensus full-year profit of CNY14.4m. "Footprint growth itself is not the main KPI," chief operating officer Bin Luo said on 27 August, naming profitable sites as the goal. Gaotu moved most of the three, up 49.5% over three months, and is still down 14.3% over twelve.

Thirty days with nothing in them

No filing falls inside the window: TAL last reported on 30 July, New Oriental on 29 July, Gaotu on 27 August. A Central Committee and State Council document on vocational education lifted all three in one session, TAL more than 4% and Gaotu about 1%. The likelier reading of the re-rating is sector and policy rather than anything that happened in a classroom. What the results do support is operating leverage at TAL, where a year-old capacity decision is now being absorbed, and the margin step at New Oriental. What they do not support is growth still accelerating anywhere in the group: New Oriental, the fastest accelerator on paper, had already eased back to a milder uptrend on 5 October.

Beijing has not softened. In January regulators fined an unregistered education firm RMB67m, explicitly to kill rumors of an academic-tutoring comeback. What these three sell is what the rules permit, collected three months at a time, and three months is all the warning their balance sheets are allowed to give.